2012 International Conference on Networks and Information (ICNI 2012)
IPCSIT vol. 57 (2012) © (2012) IACSIT Press, Singapore
DOI: 10.7763/IPCSIT.2012.V57.09
+
National Institute of Technology Raipur, Raipur, India
Clustering has been proven to be a promising approach for mimicking the operation of the fixed infrastructure and managing the resources in multi-hop networks. In order to achieve good performance, the formation and maintenance procedure of clusters should operate with minimum overhead, allowing mobile nodes to join and leave without perturbing the membership of the cluster and preserving current cluster structure as much as possible. In this paper, we propose a Flexible Weight Based Clustering Algorithm
(FWCA) influenced by Link Failure Prediction QoS Routing Protocol in Mobile Ad hoc Networks. The goals are yielding low number of clusters, maintaining stable clusters, minimizing the number of invocations for the algorithm and maximizing lifetime of mobile nodes in the system. Through simulations we have compared the performance of our algorithm with that of WCA in terms of the number of clusters formed, number of re-affiliations, number of states transitions on each clusterhead and number of clusterheads changes. The results demonstrate the superior performance of the proposed algorithm.
Election
1.
Introduction
In recent years, most research is focusing on clustering approaches for multi-hop networks because of its effectiveness in building a virtual backbone formed by a set of suitable clusterheads to guarantee the communications across clusters. An example of multi-hop networks is an ad hoc network (MANET) characterized by a collection of wireless hosts that are arbitrarily and randomly changing their locations and capabilities without the existence of any centralized entity. The main objective of clustering is to elect suitable nodes’ representatives, i.e. clusterheads (CHs) and to store minimum topology information. Each
CH will act as a temporary base station within its zone or cluster and communicates with other CHs.
Therefore, any clustering scheme should be adaptive to such changes with minimum clustering management overhead incurred by changes in the network topology. To establish a cluster, traditional clustering algorithms suggest CH election exclusively based on nodes’ IDs or location information and involve frequent broadcasting of control packets, even when network topology remains unchanged. Most recent work takes into account additional metrics (such as energy and mobility) and optimizes initial clustering. However, in many situations re-clustering procedure is hardly ever invoked; hence initially elected CHs soon waste their batteries due to serving the other members for longer periods of time. In addition, a topology control mechanism is required to mitigate the vulnerability of such clusters due to node joining/leaving and link failures. It aims to reduce interference and energy consumption, to increase the effective network capacity, and to reduce the end to end delay. Centralized algorithms rely on the assumption that the elected CH is responsible of the cluster’s maintenance. However, these algorithms suffer from single point (CH) of bottleneck especially in highly mobile environments; hence initially elected CHs have to collect excessive amounts of information and soon reach battery exhaustion. On the other hand, distributed algorithms are
+ Corresponding author. Tel.: ++91-9407627136
E-mail address : spandey96@gmail.com.
44

more adaptive to mobility due to the fact that the maintenance is done in collaboration between all the nodes where each node relies on the local information collected from the nearby nodes. Although the distributed manner is preferred for MANET, it lacks a major drawback in achieving and guarantying a strong connectivity between the nodes.
In order to simplify the maintenance, especially in high mobility scenarios, we investigate an algorithm that generates one-hop clusters. In this way, the goals of this paper are to maintain stable clusters with a lowest number of clusterheads, to minimize the number of invocations for the clustering formation/maintenance and to maximize the lifetime of mobile nodes in the system. To achieve these goals, we propose a Flexible Weight Based Clustering Algorithm (FWCA) which utilizes factors like the node degree, remaining battery power, transmission power, and node mobility for the clusterheads’ election. Our algorithm differs from others in that it is based on the clusters’ capacity and it uses the link lifetime instead of the node mobility for the maintenance procedure. We refer this to the fact that the node mobility metric does not affect the election of a CH as much as the link stability metric does. The simulations results show that the proposed algorithm provides better performance in terms of number of formed clusters, number of re-affiliations, average number of transition (state change) on CHs and number of clusterheads changes when compared to that of other weight based algorithms such as WCA. The paper is organized as follows. In
Section 2, we review several clustering algorithms proposed previously. Section 3 presents the proposed algorithm for ad hoc networks. Section 4 presents the analyzed performance of the proposed algorithm.
Finally, Section 5 concludes this paper.
2.
Previous work
A large number of approaches have been proposed for the election of clusterheads in mobile ad hoc networks. The Highest-Degree [2] uses the degree of a node as a metric for the selection of clusterheads. The degree of a node is the number of neighbors each node has. The node with maximum degree is chosen as a clusterhead; since the degree of a node changes very frequently, the CHs are not likely to play their role as clusterheads for very long. In addition, as the number of ordinary nodes in a cluster is increased, the throughput drops and system performance degrades. The Lowest- Identifier (LID) [3, 4, 5] chooses the node with the lowest ID as a clusterhead, the system performance is better than Highest-Degree in terms of throughput. However, those CHs with smaller IDs suffer from the battery drainage, resulting short lifetime of the system. The Distributed Clustering Algorithm (DCA) [6] and Distributed Mobility Adaptive clustering algorithm (DMAC) [7] are enhanced versions of LID; each node has a unique weight instead of just the node’s
ID, these weights are used for the selection of CHs. A node is chosen to be a clusterhead if its weight is higher than any of its neighbor’s weight; otherwise, it joins a neighboring clusterhead. The DCA makes an assumption that the network topology does not change during the execution of the algorithm. Thus, it is proven to be useful for static networks when the nodes either do not move or move very slowly. The
DMAC algorithm, on the other hand, adapts itself to the network topology changes and therefore can be used for any mobile networks. However, the assignment of weights has not been discussed in the both algorithms and there are no optimizations on the system parameters such as throughput and power control.
Algorithm (LCC) [9] allows minimizing clusterhead changes that occur when two CHs come into direct contact. In such a case, one of them will give up its role and some of the nodes in one cluster may not be members of the other CH’s cluster. Therefore, some nodes must become CH while causing a lot of reelections because of the propagation of such changes across the entire network. Maximum Connectivity
Clustering (MCC) [10] is based on the degree of connectivity. A node is elected as CH if it is the highest connected node. This is not suitable in dynamic network topologies where the degree of connectivity changes rapidly. The Weighted Clustering Algorithm (WCA) [11] is based on the use of a combined weight metric that takes into account several parameters like the node-degree, distances with all its neighbors, node speed and the time spent as a clusterhead. Although WCA has proved better performance than all the previous algorithms, it lacks a drawback in knowing the weights of all the nodes before starting the clustering process and in draining the CHs rapidly.
45

3.
A Flexible Weight Based Clustering Algorithm and LFPQR
In this section, we describe the fundamental concepts used to achieve the paper’s goals. First, we allocate
IDs for the nodes and the clusterheads. We use the MAC Address as the node ID in order to avoid the conflicts between IDs in the zone. Hence, the node ID (My_MAC) is unique within a cluster; the CH ID is the node ID of the CH (CH_MAC) in the cluster. The CH ID appended with the node ID forms a unique identifier for every node in the ad hoc network. Every node in the cluster will have information about its CH so that it can communicate across the cluster. Finally, the weight parameter is periodically calculated by each node in order to indicate the suitability of a node for playing cluster head’s role.
3.1.
As shown in figure 1, the goal is to build an architecture based on clusters. Every cluster has a limited number of nodes which defines its size. It also has a CH for communication across the cluster. The nodes collaborate to select the best CH. A CH must be able to manage its members, to accept or to refuse the adhesion of new arrivals based on its capacity without perturbing the functionality of the other members. Fig.
1. Example of the architecture As explained above, a ‘Counter’ is maintained by each node in order to count the number of nodes inside a cluster and to guarantee that the cluster size does not exceed a predefine ‘N’ in terms of number of nodes by cluster. Each node Ni (member or CH) is identified by a state such as: Ni
(idnode, idCH, Weight, Counter, N), it also has to maintain a ‘node_table’ wherein the information of the local members is stored. However, the CHs maintain another clusterhead information table ‘CH_table’ wherein the information about the other CHs is stored. The format of these tables is defined as: node_table
(idnode, Weight) and CH_table (idCH, Weight) .
Fig. 1. Example of the architecture
In complex networks, the nodes must coordinate between each other to update their tables. The Hello messages are used to complete this role. A Hello contains the state of the node; it is periodically exchanged either between CHs or between each CH and its members in order to update the ‘CH_table’ and the
‘node_table’ respectively
3.2.
EWCA
The clusterheads’ election is based on the weight values of the nodes. Each node computes its weight value based on the following parameters:
1.
The degree difference: defined as the difference between the cluster’s size ‘N’ and the actual number of neighbors. It allows estimating the remaining number of nodes that each node can still handle.
2.
The actual transmission power of the node.
3.
The average speed of the node.
4.
The remaining battery power of the node.
46

These parameters are inspired from those used in WCA [11], except the actual transmission power ‘Pi’ and the remaining battery power ‘Ei’. We focus on Pi instead of the sum of distance used in WCA in order to elect the node which can cover the largest range, thus, minimizing the number of clusters generated. In addition, the Ei factor is a better measure than the cumulative time during which the node acts as a CH that is used in WCA, because it allows to extend the lifetime of nodes by relinquish the role as a CH in case of insufficient battery power. Figure 2 shows steps to calculate the weight value.
The basic idea is to combine each of the above system parameters with certain weighing factors depending on the system needs. The flexibility of changing the weighting factors helps us apply our algorithm to various networks. We suppose that the nodes have the same needs. For example, in low mobility environment, we can privilege the remaining battery parameter, thus the factor ‘b’ can be made smaller for all the nodes.
Procedure for calculating the weight of node i
• Find the degree d of the nodc i by counting its neighbours
• Compute the degree difference
i
d i
N for the node i, where N is a threshold for the cluster’s size in terms of number of nodes;
• Compute the remaining battery power E i
for the node 1;
• Compute the actual transmission power P i
• Compute the average speed S i
of the node I; of the node until the current time T:
S i
T
t
T
= 1
X t
X t − 1
2
Y t
Y t − 1 where Y) is the coordinate of node i at time t:
2
• Calculate the combined weight W of the node 1:
W i
= a*A i
+ b*F i
+ c*P i
— d*S i
• where a, b, e et d are the weighting factors:
We suppose that the nodes have the same needs. For example, in low mobility environment, we can privilege the remaining battery parameter, thus the factor ‘b’ can be made smaller for all the nodes. On the other hand, we believe that the relative mobility Mi is a better factor than the average speed Si. In the first stage, we still use Si instead of Mi because it is impossible to estimate Mi when the node is alone in the zone without any other reference point (neighbors). In the second stage, the re-election is more sophisticated.
Therefore, we use the link lifetime metric which seems more realistic than the relative mobility metric to see whether it is worth to re-elect or to continue with the old CH. Initially, each node broadcasts its state (Join with idnode = My_MAC) to notify its presence to the neighbors. Each node builds its neighbors’ list based on the received states. After that, the election procedure is executed once the topology is stabled, and the node having the lowest weight is chosen as CH.
LFPQR Algorithm
When a node j receives a route request message (RREQ in AODV) from a node I, it predicts the future condition by considering power level and LLT in relation to node i. If its power level is above the threshold
(0.1 Tr) and its LLT is greater than the threshold, node j will forward this RREQ to next hop; otherwise it will drop the route request message. The same procedure is repeated for all the nodes till the destination node is reached.
In LFPQR algorithm, more stable paths are found during route discovery. Here, the stable path means, the packets which traverses on these paths will not experience more delays and improves the delivery ratio.
Hence, QoS violation is reduced. Moreover, LFPQR increases the network life time of the MANET by the second heuristic. In addition, algorithm implementation is so simple.
3.3.
Once a wireless node is activated, its idCH field is equal to NULL since it does not belong to any cluster.
The node continuously monitors the channel until it figures out that there is some activity in its
47

neighbourhood. This is due to the ability to receive the signals from other present nodes in the network. The node still has no stable state, thus its state is not full identified. In this case, it broadcasts a Join_Request in order to join the most powerful clusterhead. Thus, it waits either for a welcome_ACK or for a welcome_NACK. When the entry node does receive neither welcome_ACK nor welcome_NACK, it may increase its transmission power in order to broadcast another Join_Request that may reach the farthest clusterheads. If this persists for certain number of attempts, the node declares itself as an isolated node, readjusts its transmission power and restarts by broadcasting a new Join_Request after a period of time. We note that just the CHs may response by a welcome_ACK or welcome_NACK; the ordinary members have to ignore any Join_Request received even if they are in the transmission range of the new entry node. When the node receives multiple welcome_ACK, it selects the one which has the lowest weight. After that, it sends a
Join_Accept to the chosen clusterhead and waits for CH_ACK from this CH. The CH_ACK has to contain a confirmation that the idnode has been added to the CH_table. Thus the node can fully-define its state. The reason that we use four ways to confirm the joining procedure is to prevent other CHs that they can serve the entry node to add this node to their tables and cause conflicts.
3.4.
A CH has an idnode field is equal to idCH field. As a CH, the node calculates periodically its weight, thus it sends periodically Hello messages to its members and to the neighboring CHs in order to update the node_table and CH_table respectively. The CH must monitor the channel for
Leave, Hello and Join_Request messages. When the CH receives a Leave message, it updates the node_table and broadcasts a Hello message to its members and to its neighboring CHs to inform them that its previous ‘Counter’ was decremented. When the CH receives a Hello from a neighboring CH, it updates the
CH_table. If the Hello’s source is a node member, the CH updates the node_table and verifies its weight. In the case of a lowest weight, the CH must invoke the re-election procedure. We restrict this procedure to the
CHs in order to simplify the maintenance and the complexity of the cluster management. The re-election does not necessarily mean that a new CH must be elected even if there is a member node having a lowest weight. When the CH receives a Join_Request (idCH =NULL) from a new arrival node or a Join_Request
(full state) from a node which belongs to another cluster
It may not be possible for all the clusters to reach the cluster size N. We have tried to reduce the number of clusters by merging those that have not attained their cluster size limit. However, in order not to rapidly drain the cluster head’s power by accepting a lot of new nodes, we define thresholds which allow the clusterhead to control the number of nodes inside its cluster. When the CH receives a Join_Request, it verifies its capacity in terms of maximum number of nodes, then it verifies the ratio of power levels of the successive Join_Request messages received from the requester member, which allows getting good knowledge about the link lifetime metric between the CH and the requester node. Hence, the clusterhead does not definitively accept the merging until it is certain that the power level of the last received messages from the member is greater than the power level of the first received messages. In this way, the CH is sure that the member is moving closer to it. If not, the CH realizes that the link is going to break and it is no need to add this node in the node_table because it is going to leave soon. Finally, when a CH receives a
CH_Request from a node which desires to be CH, it must accept the request by adding the node as a new arrival CH in the topology, send a CH_Response to the node, update CH_table and broadcast a Hello message to the neighboring CHs.
It is favourable when the CH stays in the cluster for a longer time, as time need not be spent in reelection of a CH frequently. The re-election is not periodically invoked; it is performed just in case of a lowest received weight, it allows minimizing the generated overhead encountered in previous works. As we explained above, the re-election may not result a new CH, it depends on the stability of the new node for playing the CH’s role. In the case where a new CH must be elected, the procedure should be soft and flexible in order not to perturb the clusters while to copying the databases from the old CH to the new CH. We limit the execution of the algorithm where there is a CH or a network change in order not to impact the whole ad hoc topology. Thus the furthest nodes are not affected by any problem which occurs in other clusters; therefore they are up to date about the size’s changes of any cluster in the network.
48

4.
Performance Analysis
We are interested in studying the performance of the following:
- Average number of clusters: taken as stand for the quality of the cluster maintenance algorithm.
- Average transition number on each CH: defines the number of times an elected CH changes its state from
CH to a node member, thus the numbers of re-elections of a node as CH.
- Average number of CH changes: defines the number of changes occurred on the CHs during the entire simulation.
- Re-affiliation count: defines the number of different clusters a node joins during the time simulation.
4.1.
The simulations scenarios were randomly generated using our scenarios generator [12] which allows inputting parameters such as min and max speed, pause times, area, Hello interval, number of nodes, terrain configuration and the mobility model. At the physical layer, the generator uses a radio model that takes into account the path-loss, the used terrain which is a 3D environment, the frequency and the transmission range defined for the scenario. We note that the maximum radius of a mobile radio when operating at full transmission power and having an effective communication range is 300 meters which is a design parameter of some IEEE 802.11 products. The simulation parameters have been listed in table II. In the following simulation, all the nodes follow the Random Walk Mobility Model used in the scenario generator with speed ranging from 3 Km/h to 10 Km/h. In order to study the effect of the network density on the resulting topologies and to evaluate the cluster maintenance algorithm, we varied the number of the nodes inside the terrain and the power transmission range parameter. We study the stability of the ad hoc network in terms of number of formed clusters, number of clusterhead changes, number of transition on each CH, and number of re-affiliations for different transmission ranges and network densities.
Table I. Simulation Parameters
Parameter Meaning Value
N
X x Y
Speed
Number of nodes
Size of the network
Speed of the nodes
20 – 100
500m x 500m
3 – 10Km/h
R
PT
HI
Frequency
Cluster size
Duration
Transmission range
Pause Time
Hello Interval
Frequency band
Number of members
Time of simulation
30 – 300 m
0 sec
5 sec
5.4 GHZ
15 nodes
300 sec
4.2.
The number of nodes used in the simulation results varies between 20 and 100. The simulations were run for 300 seconds. The cluster size was fixed at 15. We depict some statistics on the formed clusters for different transmission ranges. In the first set of simulations, the scalability of the algorithm is measured in terms of nodes density and transmission range.
Figure 3 shows that for small ranges, most of nodes remain out of each other’s transmission range, thus the number of clusters is relatively high and the network may become disconnected because there are no other choices. When transmission range increases, more nodes can hear each other.
The average number of clusters formed decreases and the clusters become larger in size.
49

Fig. 2 Transmission range vs. Avg. Number of Clusters
In Figure 4, we note that the performance difference is small between WCA and FWCA with respect to the average number of clusters. This is because both algorithms are variations of a local weight based clustering technique that forms one-hop clusters. For high transmission range (more than 250 m),
WCA generates less CH than FWCA but to the detriment of a large number of transitions on each CH, where the stability is one of the important criteria in clustering because the frequent changes of CH adversely affect the performance of the clustering algorithm. As a result, our algorithm gives better performance in terms of stability when the node density in the network is high.
Fig. 3: Number of Nodes vs. Number of Cluster
The result of the average number of re-affiliations due to increasing node density is depicted in figure 5.
For a transmission range of 120 meters, the number of re-affiliations increased when varying the number of nodes in the network for both our algorithm and WCA. As the number of nodes increased, the increasing rate of re-affiliations slowed down in FWCA, which was not the case in WCA.
Fig. 4: Number of Nodes vs. Re-affiliation count
50

For a node speed varying between 3 and 10 km/h, when there were 20 nodes in the network and for the same transmission range (120 m), the proposed algorithm produced 61.5% less re-affiliations than WCA.
When the number of nodes was increased to 100, our algorithm gave 66.5% less re-affiliations than WCA for the same node speed.
Packet delivery ratio is the ratio of the number of data packets reached the destination to the number of packets generated by the CBR source. LFPQR has the better delivery ratios as shown Figure 6, because it selects more stable nodes which contain better power levels. The packet loss is less due to the low probability of node failures and network partitions in LFPQR routing. This shows that improvement in routing performance.
Fig 5 Average Delay
5.
Conclusion and Future Work
This paper has presented a Flexible Weight Based Clustering Algorithm in Mobile Ad hoc Networks. It has the flexibility of assigning different weights and takes into account a combined metrics to form clusters automatically. Limiting the number of nodes inside a cluster allows restricting the number of nodes catered by a clusterhead so that it does not degrade the MAC functioning. For a fixed clusterhead election scheme, a clusterhead with constrained energy may drain its battery quickly due to heavy utilization. In order to spread the energy usage over the network and achieve a better load balancing among clusterheads, re-election of the clusterheads may be a useful strategy; the algorithm is executed only when there is a demand. Also, if a node is moving away from the clusterhead, then the algorithm is flexible and cheap enough to be applied iteratively as the network configuration changes. Therefore, such approach provides a reliable method of cluster organization for wireless ad hoc networks. Simulation results indicated that the model agrees well with the behaviour of the algorithm. Eventually, the route between two nodes changes constantly as the clusterhead set changes. We are planning to study the overhead generated by FWCA in order to evaluate its impact on the network and to complete the inter-clusters communications in future works.
6.
References
[1] Das B., Bharghavan V., "Routing in ad-hoc networks using minimum connected dominating sets," IEEE
International Conference on Communications (ICC Montreal 97), vol. 1, Jun. 1997, pp. 376-380.
[2] Gerla M., Tsai J. T. C., "Multicluster, Mobile, Multimedia Radio Network," ACM/Baltzer Wireless Networks
Journal 95, vol. 1, Oct. 1995, pp. 255-265.
[3] Baker D.J., Ephremides A., "A distributed algorithm for organizing mobile radio telecommunication networks,"
Proceedings of the 2 nd International Conference on Distributed Computer Systems, Apr. 1981, pp. 476-483.
[4] Baker D.J., Ephremides A., "The architectural organization of a mobile radio network via a distributed algorithm,"
IEEE Transactions on Communications, Nov. 1981, pp. 1694-1701.
[5] Ephremides A., Wieselthier J. E., Baker D. J., "A Design Concept for Reliable Mobile Radio Networks with
Frequency Hopping Signaling," Proceedings of the IEEE, vol. 75, no. 1, Jan. 1987, pp. 56-73.
51

[6] Basagni S., "Distributed clustering for ad hoc networks," Proceedings of International Symposium on Parallel
Architectures, Algorithms and Networks, Jun. 1999, pp. 310-315.
[7] Basagni S., "Distributed and mobility-adaptive clustering for multimedia support in multi-hop wireless networks,"
Proceedings of Vehicular Technology Conference, VTC, vol. 2, fall 1999, pp. 889-893.
[8] Basu P., Kham N., Little T. D. C., "A mobility based metric for clustering in mobile ad hoc networks," International
Conference on Distributed Computing Systems Workshop, Apr. 2001.
[9] Chiang C. C., Wu H. K., Liu W., Gerla M., "Routing in Clustered Multihop, Mobile Wireless Networks with
Fading Channel," IEEE SICON, Singapore, Apr. 1997.
[10] Parekh A.K., "Selecting routers in ad-hoc wireless networks," Proceedings of the SBT/IEEE International
Telecommunications Symposium, Aug. 1994.
[11] Chatterjee M., Das S.K., Turgut D., "WCA: A Weighted Clustering Algorithm for Mobile Ad Hoc Networks,"
Cluster Computing Journal, vol. 5, no. 2, Apr. 2002, pp. 193-204.
[12] Agba L., Gagnon F., Kouki A., "Scenarios generator for ad hoc networks," International Symposium on Industrial
Electronics, Montreal, Canada, Jul. 2006.
52